Scholarlink

<p>Sentiment Analysis (SA) is pivotal for extracting people’s opinions and feedback from texts on specific topics or products. Arabic texts often present unique challenges for SA due to their intricate combination of multiple dialects alongside Modern Standard Arabic (MSA), leading to complex regional variations. This paper addresses these complexities by introducing a sentiment analysis system tailored for the travel domain. Our system is equipped to analyze sentences encompassing a blend of Arabic dialects from regions such as the Maghreb, Nile Basin, Levant, Gulf, and Yemen, spanning 25 cities. Initially, sentences undergo preprocessing and tokenization. To account for potential misspellings, we developed a probability-driven approach to generate word variations. These variations are then processed by a trained Long Short-Term Memory Recurrent Neural Network (LSTM RNN) to predict their MSA synonyms. Concurrently, a comprehensive ontology is employed, which not only extracts the relevant dialectal variation mapped to the predicted MSA using similarity measurements but also assigns the appropriate sentiment class. The ontology further enhances accuracy by addressing excessive words and negations, given their significant role in sentiment determination. Our approach showcases its robustness with promising experimental results, achieving an overall accuracy of 85%. This work signifies a notable advancement in sentiment analysis for multi-dialectal Arabic texts, bridging the gap between diverse regional nuances and effective opinion mining.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Hybrid LSTM RNN and Ontology-Based Approach to Sentiment Analysis Across Multiple Arabic Dialects

  • 期刊论文

  • 发表日期：

    2025年7月30日

查看全文
  • Rahima Bentrcia,
  • Kaddour Djakam,
  • Okba Madani,
  • Ashraf Elnagar

摘要

Sentiment Analysis (SA) is pivotal for extracting people’s opinions and feedback from texts on specific topics or products. Arabic texts often present unique challenges for SA due to their intricate combination of multiple dialects alongside Modern Standard Arabic (MSA), leading to complex regional variations. This paper addresses these complexities by introducing a sentiment analysis system tailored for the travel domain. Our system is equipped to analyze sentences encompassing a blend of Arabic dialects from regions such as the Maghreb, Nile Basin, Levant, Gulf, and Yemen, spanning 25 cities. Initially, sentences undergo preprocessing and tokenization. To account for potential misspellings, we developed a probability-driven approach to generate word variations. These variations are then processed by a trained Long Short-Term Memory Recurrent Neural Network (LSTM RNN) to predict their MSA synonyms. Concurrently, a comprehensive ontology is employed, which not only extracts the relevant dialectal variation mapped to the predicted MSA using similarity measurements but also assigns the appropriate sentiment class. The ontology further enhances accuracy by addressing excessive words and negations, given their significant role in sentiment determination. Our approach showcases its robustness with promising experimental results, achieving an overall accuracy of 85%. This work signifies a notable advancement in sentiment analysis for multi-dialectal Arabic texts, bridging the gap between diverse regional nuances and effective opinion mining.